Konstantin Zlobin: AI struggles when planning pilots meet operational reality
Why scaling AI in integrated planning depends on data, governance and decision ownership
Konstantin Zlobin reflected on a discussion at LogiPharma Asia 2026 around a central planning question: how can companies move AI beyond pilots and embed it in planning systems that can scale?
His key observation is that AI itself is not necessarily the main constraint. Assistive AI is already improving planner productivity, but scaling it inside Integrated Planning requires the technology to support the complete decision process rather than isolated tasks. Demand and supply decisions must remain connected, with consistent data, structured planning cycles and clear governance.
This distinction matters because a pilot can operate with curated data, a narrow use case and close expert supervision. Operational planning is different. It involves changing assumptions, competing objectives, multiple planning horizons and decisions that affect inventory, capacity, service, cost and regulatory commitments.
Zlobin’s formulation is useful: AI often does not fail in the pilot; it struggles when it encounters operational reality. At scale, Data Governance, system integration and decision ownership become as important as model performance.
The pharma context makes this especially relevant. LogiPharma Asia brings together pharmaceutical and life-science supply chain, operations and logistics leaders, with a focus on resilience, collaboration and emerging technologies. The 2026 edition took place in Singapore on June 16–17.
Dataleo angle
This is a strong Supply Chain AI signal because it separates controlled experimentation from industrialized decision support. The critical question is not whether an assistant can generate forecast commentary or summarize an exception. It is whether AI can operate inside a governed planning cycle with trusted inputs, explicit decision rights and human override.
Before scaling, planning leaders should define which decision is being improved, who owns the underlying logic, how outputs enter S&OP and demand-supply workflows, and what happens when the recommendation is wrong. Without those controls, AI risks becoming another isolated layer beside APS, ERP and spreadsheets rather than part of a coherent Decision Architecture.